Research On Target Area Segmentation Method For Image Monitoring Of Industrial Smoke And Dust Emission

Air pollution threatens the living environment of human beings gradually in recent years,and industrial smoke is the major source of air pollution.It is particularly important how to monitor industrial smoke effectively in real-time.And Ringelman emittance coefficient is an effective way to describe smoke blackness.The existing methods detect industrial smoke images in daytime environment by manual observation generally.Comparing with Ringelman emittance image to judge the darkness level.It exists low efficiency or a point-of-view situation.Some scholars proposed using digital image processing methods to monitor industrial smoke images.The difficulty of this method is how to separate the smoke target areas from the background accurately.According to this problem,this paper studies industrial smoke images segmentation method during the daytime period mainly,which is described as follows:(1)Industrial smoke image segmentation method based on background modeling and feature matching.In order to separate the smoke target areas from the background accurately.Firstly,building a background update model,using the method of differential accumulation to divide the smoke target areas.Filling the hole area with morphological filling method,and it obtained a rough segmentation area.Finally,using feature matching method to fine segmentation.This method was used to segment the smoke video images under five different scenes,and a variety of industrial smoke images segmentation methods were compared.The experimental results show that this method is superior to other methods in the accuracy of recall and investigation.(2)Industrial smoke images segmentation method based on convolutional neural network of transfer learning.In order to excavate further depth characteristics of smoke images and improve the accuracy of segmentation.Firstly,constructing convolutional neural network model for industrial soot image segmentation.Using the transfer learning method,the VGG16 model is fine tuned with the industrial sootimage data set produced in this section.It gets applicable to the new network model of industrial smoke image segmentation in this paper.Finally,making segmentation experiment of five different industrial smoke scenes images.Experimental results show that the recall and precision of this method are higher than the method in the previous section.(3)Industrial smoke images segmentation method based on full convolutional network.In order to perform pixel-level segmentation on industrial smoke images.Firstly,constructing a full convolutional network model of industrial smoke images.Preprocessing industrial smoke data sets and inputting into a full convolutional network model,and iterative training is conducted.And then obtaining industrial smoke image segmentation network model,which is suitable for in this paper.Finally,it will do segmentation experiments.Experimental results show that the recall and precision of this method are higher than the previous two methods,and achieve pixel-level segmentation.